Sentiment Analysis - Game of Thrones
Group-10¶
- Dhawal Shah
- Eshan Kulkarni
- Leena Patil
- Prasanna Patil
- Pritam Bhatkar
- Wrujuta Dhawale
Introduction of the topic and dataset¶
- Summers span decades. Winters can last a lifetime. And the struggle for the Iron Throne begins. Based on the bestselling book series by
George R.R. Martin and created by David Benioff and D.B. Weiss.
Introduction to the dataset¶
Source: Twitter
user_id - unique twitter_id of every user
- created_at - date & time tweets where tweeted
- source - source of tweets
- text - actual tweets
- display_text_width - length of tweets
- geo_coords - geo location of tweeets
Objective¶
- To perform Sentiment Analysis on Game of Thrones, based on speculations & reactions from US twitter users.
Tools & Technologies used¶
- Python
- NLTK libraries
- Spacy
- Wordcloud libraries
- Scikit-learn (Topic Modelling)
- VADER (Sentiment Analysis)
- Tweepy (Twitter API)
Project Workflow¶
- Data Acquisition
- Prepare and Clean Data
- Explore Data
- Data Analysis
- Visualize results
Import libraries¶
Setup twitter API for Web Scraping¶
Use twitter API to scrape tweets for Game of Thrones¶
Check number of rows and columns in dataset¶
In [5]:
df_got.shape
Out[5]:
- Dataset contains columns which have missing values and thus needs data cleaning
In [7]:
df_got.head()
Out[7]:
Data Cleaning & Preparation¶
In [31]:
df_got.info()
In [19]:
df_got.head()
Out[19]:
Data Preprocessing¶
Apply preprocessing functions to the tweets column¶
Get compound score & labels of tweets using VADER¶
Get text data of wordcloud for each episode¶
Exploratory Data Analysis¶
In [50]:
df_got.info()
In [51]:
df_got.describe()
Out[51]:
In [52]:
df_got.head()
Out[52]:
Positive Percent¶
In [53]:
pos = df_got[df_got['tweets_compound_score'] > 0.0]
pos_per = (len(pos)/len(df_got)) * 100
pos_per
Out[53]:
Neutral Percent¶
In [54]:
neu = df_got[df_got['tweets_compound_score'] == 0.0]
neu_per = (len(neu)/len(df_got)) * 100
neu_per
Out[54]:
Negative Percent¶
In [55]:
neg = df_got[df_got['tweets_compound_score'] < 0.0]
neg_per = (len(neg)/len(df_got)) * 100
neg_per
Out[55]:
Character Density distribution¶
In [56]:
plt.figure(figsize=[10, 10])
chr_length = df_got['character_length']
sns.distplot(chr_length, rug=True);
Compound Score Density distribution¶
In [57]:
plt.figure(figsize=[10, 10])
cpd_score = df_got['tweets_compound_score']
sns.distplot(cpd_score);
Topic Modelling for Episode 5¶
In [59]:
episode_5.info()
In [60]:
episode_5.head()
Out[60]:
In [71]:
topics = ldamodel.print_topics(num_words=10)
for topic in topics:
print(topic)
Visualization¶
Sentiment Score¶
In [74]:
plt.figure(figsize=(30, 20))
plt.ylim(-0.025, 0.14)
plt.xlabel('Timeline', fontsize=30)
plt.ylabel('Compound Score', fontsize=30)
plot_score.plot(kind='line', color='b', alpha=0.5)
plt.grid()
plt.show();
Winterfell¶
In [75]:
text = winterfell(x)
plt.figure(figsize=(20, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'go', 'see', 'season', 'think', 'one', 'watch', 'fan', 'time', 'start', 'maybe',
'scene', 'anyone', 'really', 'still', 'anyone', 'already', 'anything', 'never', 'everyone', 'even'])
wc = WordCloud(stopwords=stopwords, background_color='black')
wc.generate(text)
# Display the generated image:
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
plt.show()
A knight of Seven Kingdoms¶
In [76]:
text = seven_kingdoms(x)
plt.figure(figsize=(20, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'go', 'start', 'show', 'time', 'fan', 'new', 'see', 'know', 'watch', 'one', 'think', 'tonight',
'say', 'good', 'want', 'still', 'end', 'already', 'anything', 'never', 'everyone', 'even'])
wc = WordCloud(stopwords=stopwords, background_color='black')
wc.generate(text)
# Display the generated image:
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
plt.show()
The Long Night¶
In [77]:
text = long_knight(x)
plt.figure(figsize=(20, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'go', 'final', 'twitter', 'tonight', 'think', 'time', 'show', 'good', 'end', 'new', 'come',
'fan', 'start', 'know', 'one', 'say', 'spoiler', 'best', 'need', 'want', 'talk', 'see', 'today', 'last', 'catch'])
wc = WordCloud(stopwords=stopwords, contour_width=2, background_color='black')
wc.generate(text)
# Display the generated image:
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
plt.show()
The Last of Starks¶
In [78]:
text = last_of_starks(x)
plt.figure(figsize=(20, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'go', 'final', 'tonight', 'time', 'season', 'see', 'say', 'start', 'watch', 'via','make',
'show','new', 'anyone', 'really', 'still', 'anyone', 'already', 'anything', 'never', 'everyone', 'even', 'one', 'end'])
d = getcwd()
cup_mask = np.array(Image.open(path.join(d, "cup.jpg")))
wc = WordCloud(stopwords=stopwords, mask=cup_mask, contour_width=2, contour_color='grey', background_color='black')
wc.generate(text)
wc.to_file("img/cup.jpg")
image_color = ImageColorGenerator(cup_mask)
# Display the generated image:
plt.imshow(wc.recolor(color_func=image_color), interpolation='bilinear')
plt.axis("off")
plt.show()
The Bells¶
In [79]:
text = the_bells(x)
plt.figure(figsize=(20, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'new', 'petit', 'via', 'one', 'season', 'fan', 'time', 'theory', 'last',
'see', 'tonight', 'good', 'think', 'say', 'want', 'show', 'know', 'watch', 'go', 'dont'])
d = getcwd()
bell_mask = np.array(Image.open(path.join(d, "bell.png")))
wc = WordCloud(stopwords=stopwords, mask=bell_mask, contour_width=2, contour_color='orange')
wc.generate(text)
wc.to_file("img/bell.png")
#bell_mask = ImageColorGenerator(bell_mask)
# Display the generated image:
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
plt.show()
The Iron Throne¶
In [63]:
text = iron_throne(x)
plt.figure(figsize=(15, 10))
# Create and generate a word cloud image:
stopwords = set(STOPWORDS)
stopwords.update(['throne', 'game', 'new', 'petit', 'via', 'one', 'season', 'fan', 'time', 'theory', 'last',
'see', 'tonight', 'good', 'think', 'say', 'want', 'show', 'know', 'watch', 'go', 'dont'])
d = getcwd()
throne_mask = np.array(Image.open(path.join(d, "throne.jpg")))
wc = WordCloud(stopwords=stopwords, mask=throne_mask, contour_width=2, contour_color='firebrick', background_color='black')
wc.generate(text)
throne_mask = ImageColorGenerator(throne_mask)
# Display the generated image:
plt.imshow(wc, interpolation='bilinear')
plt.axis("off")
plt.show()
Conclusion¶
- The Initial episodes had a positive reaction.
- The negative sentiment grows after the initial episodes.
- Curious topic comes with the Bells episode, which was the most controversial one, with Dany seemingly deciding to burn down a city on a whim.
- Most negative sentiment observed for Episode 5 - 'The Bells'.
- Overall we have learned that the posts on social media are difficult to interpret as they are full of emotion and sarcasm.
In [ ]: